OpenAI, Broadcom to Build 10GW AI Accelerator Infrastructure

OpenAI and Broadcom are launching a strategic collaboration to develop and deploy 10GW of AI accelerators. This multi-year agreement tasks OpenAI with designing the hardware systems, while Broadcom will manage their development and implementation to meet the growing global demand for AI computing capacity.

"Broadcom’s collaboration with OpenAI signifies a pivotal moment in the pursuit of artificial general intelligence," says Hock Tan, President and CEO, Broadcom. This strategic alliance integrates OpenAI's knowledge from developing frontier models directly into the hardware design to optimize system performance and efficiency.

The announcement addresses unprecedented demand for specialized AI infrastructure, as traditional hardware architectures face limitations in handling the scale and complexity of next-generation language models. To overcome these challenges, leading organizations are opting for custom silicon designs, known as custom accelerators, that allow for deep optimization between software and hardware.

By designing its own accelerators, OpenAI plans to incorporate what it has learned from creating models and products directly into the chip architecture. This vertical integration has the potential to unlock new levels of capability and efficiency, reducing latencies and operational costs at scale. The decision to use an Ethernet-based architecture for scaling AI clusters underscores an industry trend toward more open, scalable and cost-effective solutions for AI data centers.

The agreement is an extension of previous co-development and supply arrangements between the two companies. The terms establish a clear implementation timeline: deployment of the racks is scheduled to begin in the second half of 2026 and is expected to conclude by the end of 2029. These systems, which will incorporate both OpenAI-designed accelerators and Broadcom networking solutions, will be installed in OpenAI facilities and its partners’ data centers.

The racks will include Broadcom’s end-to-end portfolio of connectivity solutions, covering Ethernet, PCIe, and optical solutions. "Custom accelerators combine remarkably well with standards-based Ethernet scale-up and scale-out networking solutions to provide cost and performance optimized next generation AI infrastructure," says Charlie Kawwas, President, Semiconductor Solutions Group, Broadcom.

For Broadcom, this collaboration reinforces its position in the AI infrastructure market and validates Ethernet as a leading technology for networking in large-scale AI data centers.